On Wed, 21 Sep 2005, Peter Troeger wrote:
Hi,
I am stumbling over the first issues while improving the Condor DRMAA implementation ;-)
In one test case, the SUN DRMAA test suite calls drmaa_wait with NULL as value for the rusage parameter. The Condor implementation currently treats this as an error, since there is no statement in the specs that this parameter can be optional. For SGE this seems to be ok, it simply does not return rusage data in this case. What do we declare as correct behavior ?
Hi Peter, I consider Grid Engine behavoir reasonable as it is borrowed from POSIX wait3() If rusage is not a null pointer, a summary of the resources used by the terminated process and all its children is returned. Only the user time used and the system time used are currently available. They are returned in the ru_utime and ru_stime, members of the rusage structure, respectively. in case the DRMAA application doesn't care about resource usage enforcing non-NULL rusage be passed to drmaa_wait() would make the application needlessly more voluminous, since the drmaa_attr_values_t rusage array would need to be released via drmaa_release_attr_values() even though it isn't used.
Regards, Peter.
P.S.: I would be happy if someone could take the responsibility for opening according tracker items, at least for the next several weeks. Thanks.
I encountered DRMAA C binding GWD-R document doesn't state explicitely rusage may be NULL. I added it as tracker item. Regards, Andreas